Deep Learning Based Enhanced stroke prediction on Deep Resnet Convolution neural network using optimized feature selection and classification

Abstract

Nowadays heart disease is most dangerous to create stroke suddenly attain heart attack leads to death. By increasing fast foods, genetic life cycle and age factors are the important reason for this disease which affects the coronary heart valves to block the blood flow. The development of medical science and engineering develops computer aided design based on image analysis with support of angiogram scans to identify the disease. Most of the prevailing techniques in machine learning models are failed to identify the disease properties in terms of feature analysis to identify the exact region of the block. Due to unidentified segmentation region and higher intensity, the identification accuracy is low due to higher false positives.             By addressing the problems, to improve the stroke prediction using support vector quantization feature selection (SVQFS) based on Resnet50 convolution neural network (Resnet50-CNN). Initially the preprocessing is carried out by Min max -Neighbor vector normalization (Min-Max-NVN) to remove the noise and verify the actual margin in heart disease dataset. The cardiac stroke impact margin rate (CSIR) is analyzed by identifying the actual affected features with support of mean covariance scalar estimation. Using the support vector quantization algorithm (SVQA), to find the actual features to select the important margins to reduce the feature scaling and dimension of non-scaling features. Then resnet50 –CNN is applied to predict the feature scaling with soft-max activation function and effectively predict the disease class to categorize normal and abnormal. The proposed system effectively segment the stroke region by identifying the relevance feature margins and support vector dimension probability despond on disease impact rate to improve the detection accuracy. Also the performance increasing higher recall, precision rate by combative best rue positives feature limits with low false rate compared to the existing system.